Lead Investigator: Angela Wu, University of Oxford
Title of Proposal Research: Investigating the association between smoking cessation and mental health in people with and without psychiatric disorders
Vivli Data Request: 6544
Funding Source: Angela D Wu is funded by Nuffield Department of Primary Care Health Sciences through a grant received by the British Heart Foundation. Prof. Paul Aveyard is an NIHR senior investigator and funded by NIHR Oxford Biomedical Research Centre and NIHR Oxford Applied Research Collaboration. Dr. Gemma Taylor is funded by a Cancer Research UK Postdoctoral Fellowship (C56067/A21330).
Potential Conflicts of Interest: None
Summary of the Proposed Research:
In many countries, such as the United Kingdom, the prevalence of smoking in the general population has decreased from approximately 46% in 1970s to 14.9% in 2018. The decrease in prevalence is not to the same extent among people with mental health conditions, with the prevalence remaining at around 40% from 1993 to 2013. Many individuals who smoke say that they wish to quit; however, many continue to believe that smoking provides them with mental health benefits. This belief is shared among individuals with and without psychiatric disorders. Smokers, therefore, may be less likely to cease smoking if they believe that their mental health will decline as a result. Additionally, health professionals may also thereby be reluctant to suggest smoking cessation to certain smokers due to impact on mental health. However, there is evidence suggesting that there is an association between stopping smoking and improved mental health. Given that the rate of smoking in individuals diagnosed with psychiatric disorders is not subsiding to the same level as the general population, it is, therefore, essential to update the evidence exploring the association between mental health and smoking for populations with and without psychiatric disorders.
This project’s focus is to assess the strength of the relationship between quitting smoking and increased positive mental health for people with and without psychiatric disorders. It is not feasible to randomly assign participants to continue or quit smoking; therefore, to study the association between smoking cessation on mental health, observational analysis strategies must be used. The primary issue with traditional observational epidemiology is teasing out whether associations are causal. This study will have a unique opportunity to use three analytical approaches in a large dataset with populations with and without psychiatric disorders in order to derive results with higher causal inference confidence. We will triangulate results derived from statistical methods that differ in their ability to produce causal estimates: multivariable regression models, propensity score-adjusted models, and instrumental variable regressions.
Statistical Analysis Plan:
Primary analysis – The primary analysis will focus on results from the Hospital Anxiety and Depression Scale (HADS )(Zigmond & Snaith, 1983). Participants self-reported anxiety and depression severity. The HADS consists of fourteen individual item responses ranging in increasing severity from 0 (normal) to 3 (most severe) for a total range of 0 to 42. 7 items assess anxiety and seven assess depression, which therefore provides two subscales with ranges of 0 to 21. In each subscale, 0 to 7 is considered normal, while 15 to 21 represents severe symptoms. We will repeat analyses for the subscales for anxiety and depression.
Secondary analysis – The secondary analysis will consist of investigating the change in reported frequency and severity of suicidality and neuropsychiatric adverse events. We will also assess the incidence of mood disorders using odd ratios.
To investigate the effects of smoking cessation on mental health, we propose to use three stages of analysis: conventional linear regression, propensity score regression and instrumental variable analysis. We will also examine the dataset for potential useful repeated measures variable to adjust for time-varying measures.
Multivariable linear regression modelling
In our first analysis, we will use conventional multivariable linear regression modelling to assess the relationship between smoking cessation and change in HADS score from baseline to follow-up. All covariates listed above will be adjusted for in the model. Smoking cessation will be treated as a dummy variable (quitting=1, continuing smoking=0). Due to regression to the mean when using within-person, repeated measures data, participants’ mean change scores will not be used to measure the change in HADS score from baseline to follow-up; instead, we will use 24-week follow-up HADS scores, with adjustment for baseline HADS score (Vickers & Altman, 2001). Associations will be reported after being adjusted for basic confounders such as age and gender and other confounding factors identified.
Propensity score matching approach
In our second analysis, we will construct propensity scores using logistic regression in this model whereby the predictors are the covariates, and the outcome is smoking cessation (Glynn et al., 2006; Rosenbaum & Rubin, 1983, 1984). All baseline variables recorded in the trial will be included in the model. Therefore, each participant’s propensity score will be their conditional probability (odds) of smoking cessation. Individuals who have quit smoking will be matched to a continuing smoker with the closest propensity score on a ratio of 1:1 using the nearest neighbour greedy algorithm with no replacement, and matching will be restricted to the common support region (Taylor et al., 2015). We will subsequently estimate the association of the outcomes through an adjusted linear regression model for the propensity scores, where the exposure is smoking cessation and outcome is mental health.
Instrumental Variable Analysis
In our third analysis, the randomisation of participants into treatment or placebo groups will be used as an instrumental variable, or a proxy for our exposure (quitting versus continuing to smoke), to measure the causal relationship between smoking cessation and mental health provided that there is sufficient power. Risk differences will be reported in the outcomes through the use of additive structural mean models estimated via the generalised method of moments (Clarke & Windmeijer, 2010; Clarke & Windmeijer, 2012; Davies et al., 2017; Hansen & Singleton, 1982). We will use the Cragg-Donald Wald F statistic and the Hausman test for endogeneity to test for weak instrument bias (Hahn & Hausman, 2002; Stock & Yogo, 2005).
Model adequacy checks
The propensity score model will be checked to ensure that there is a balance of means and variances for covariates in the propensity score-matched sample (Thoemmes & Kim, 2011). We will use standard procedures to compare the relative bias of linear regression and instrumental variable methods. This procedure will consist of comparing the association between exposure and baseline covariates, and the association between the instrument and covariates (Jackson & Swanson, 2015).
For the baseline and outcome data, we will use multivariable multiple imputation to impute data for patients missing values (Royston & White, 2011). The imputation procedure will produce twenty imputed datasets, and the imputation model will include all baseline covariates (Royston, 2004).
For missing exposure data, smoking status, we will assume that those with missing data are continuing smokers. This method produces similar results as multivariable multiple imputation (Taylor et al., 2017).
We will present estimates derived from the complete case and imputed models.
We will compare the effect estimates derived from complete cases, and compare this to effect estimates derived from complete cases with the addition of imputed data
Evaluating The Safety And Efficacy Of Varenicline and Bupropion For Smoking Cessation In Subjects With And Without A History Of Psychiatric Disorders (EAGLES)
Data Contributor: Pfizer
Study ID: NCT01456936
Wu, A.D., Gao, M., Aveyard, P. and Taylor, G., 2023. Smoking Cessation and Changes in Anxiety and Depression in Adults With and Without Psychiatric Disorders. JAMA Network Open, 6(5), pp.e2316111-e2316111. Doi: 10.1001/jamanetworkopen.2023.16111